Processing method, device and equipment for operation configuration of energy storage power station and storage medium

Abstract

The invention provides an energy storage power station operation configuration processing method and device, equipment and a storage medium, and relates to the technical field of energy storage control. According to the scheme, training process data of a trained configuration analysis model are obtained; quantifying importance of each operation characteristic parameter to a model output safety result based on the data, and determining a target characteristic parameter and a weight of the target characteristic parameter; obtaining a plurality of candidate operation configurations of the target energy storage power station, performing weighted aggregation on the relative distance between each target characteristic parameter in each candidate operation configuration and the expected operation configuration based on the weight, and determining the fitness of each candidate operation configuration; determining a target operation configuration with the maximum fitting degree from the candidate operation configurations; the target operation configuration is used for configuring the target energy storage power station, and the safety state of the configured target energy storage power station is related to the target operation configuration. The accuracy of operation configuration decision of the energy storage power station can be improved, so that the safety performance of the energy storage power station is improved.

Description

Technical Field

[0001] This application relates to the field of energy storage control technology, and in particular to processing methods, apparatus, equipment, computer-readable storage media, and computer program products for the operation configuration of energy storage power stations. Background Technology

[0002] Energy storage power stations are key facilities supporting the stable operation of new power systems and the absorption of renewable energy, and determining safe operating configurations for them is a prerequisite for ensuring the safety of the power grid and the power station itself.

[0003] Currently, the decision-making process for the operation and configuration of energy storage power stations mainly relies on experience gained from historical practices to determine evaluation indicators for the operation and configuration of energy storage power stations, and then makes decisions on the operation and configuration based on these evaluation indicators.

[0004] However, due to the mixed nature of historical data and the variability of operating conditions, the evaluation indicators established based on these indicators often incorporate too much noise and subjective judgment. This can lead to a disconnect between the evaluation results of operational decisions based on these indicators and the actual safety state that the energy storage power station should achieve. Consequently, the accuracy of operational configuration decisions is reduced, making it difficult to guarantee the safety of the energy storage power station. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, and storage medium for processing the operation configuration of an energy storage power station, which can improve the accuracy of the operation configuration decision for the energy storage power station, thereby selecting a target operation configuration that is more closely matched to the desired operation configuration for the energy storage power station, and thus improving its safety performance.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, a method for processing the operation configuration of an energy storage power station is provided, the method comprising: First, the model training process data of the configuration analysis model is obtained. This model has the ability to output the safety status result of the energy storage power station based on multiple feature parameters in the energy storage power station's operational configuration. The configuration analysis model includes multiple base learners, each containing multiple nodes. The model training process data includes the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner during feature splitting learning based on the feature parameters during model training. Second, based on the model training process data, the importance quantification result of each feature parameter is determined. This importance quantification result characterizes the degree of influence of the corresponding feature parameter on the output safety status result of the energy storage power station from the configuration analysis model. Then, based on the importance quantification result, target feature parameters are selected from the feature parameters, and a first weight is determined for each target feature parameter. The target feature parameter is the feature parameter whose importance quantification result satisfies a preset importance condition. The first weight is positively correlated with the degree of influence represented by the importance quantification result of the target feature parameter. Then, multiple candidate operating configurations for the target energy storage power station are obtained. Based on the first weight of each target feature parameter, the relative distances between each target feature parameter in each candidate operating configuration and the desired operating configuration are weighted and aggregated to determine the fit degree of each candidate operating configuration. The desired operating configuration is obtained based on the target feature parameters of multiple candidate operating configurations. The relative distances to the desired operating configuration are obtained by processing the target feature parameters of each candidate operating configuration using an approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate operating configuration and the desired operating configuration. The candidate operating configuration with the highest fit degree is selected from the candidate operating configurations and used as the target operating configuration for the target energy storage power station. The target operating configuration is used to configure the target energy storage power station, and the safety state of the configured target energy storage power station is related to the target operating configuration.

[0007] This application provides a method for processing the operation configuration of an energy storage power station. First, by analyzing the Gini gain value generated by feature splitting at each node during model training, and the contribution weight of each base learner in the ensemble model, the method utilizes the feature selection and combination capabilities automatically performed by the model during training to improve output accuracy. This allows for reverse reasoning of the implicit decision-making logic within the model to quantify the importance of each feature parameter to the final safety state result. Subsequently, the feature parameters are screened for importance, and the normalized weights of the target feature parameters are determined. Compared to related methods, this application can objectively screen and weight feature parameters (evaluation indicators) based on the inherent objective laws of the data learned during model training, reducing the disconnect between the evaluation results and the noise indicators introduced by relying on historical experience. This constructs a feature evaluation system that more accurately reflects the actual safety state of the energy storage power station. Furthermore, when evaluating and deciding on candidate operation configurations for the energy storage power station, the relative distance between each candidate operation configuration and the desired operation configuration (defined by positive and negative ideal solutions) is calculated using an approximation ideal solution method, and the weights of the aforementioned target feature parameters are introduced to weighted aggregate these distances. This mechanism ensures that key characteristic parameters significantly impacting the safety status of energy storage power stations are objectively and reasonably emphasized during the evaluation of candidate operating configurations. This allows the final fitted score to more accurately assess the true effectiveness of each candidate operating configuration, thus mitigating the risk of discrepancies between evaluation results and actual effects in related technologies. Finally, by selecting the configuration with the highest fitted score as the target operating configuration, a complete decision-making loop is achieved, from objective data pattern mining to key parameter weight determination, and then to accurate evaluation and automatic optimization of multiple schemes. This improves the accuracy of energy storage power station operation decisions. It provides a direct and reliable basis for the daily operation and safety strategy formulation of energy storage power stations, enhancing their safety performance.

[0008] In one possible implementation of the first aspect, the characteristic parameters include at least one of the following dimensions: DC-side battery pack grounding characteristic dimension, power conversion characteristic dimension, relay protection characteristic dimension, and fault characteristic dimension. The DC-side battery pack grounding characteristic dimension includes: DC battery pack connection method parameters and grounding resistance value parameters. The power conversion characteristic dimension includes: converter operating condition parameters, converter performance parameters, and converter safety protection function parameters. The relay protection characteristic dimension includes: relay protection action criterion type parameters, relay protection area range parameters, relay protection device model parameters, and relay protection function parameters. The fault characteristic dimension includes: inter-pole fault current peak parameters, polar fault current peak parameters, positive and negative DC bus voltage peak parameters, DC energy storage capacitor fault voltage parameters, AC side three-phase current offset parameters, polar fault current response time parameters, undervoltage protection response time parameters, and converter blocking response time parameters.

[0009] It should be understood that the characteristic parameters in this scheme include electrical characteristic dimensions that affect the overall safety of the energy storage power station. By concretizing the abstract operation configuration into a quantifiable parameter system, it provides a comprehensive and engineering-realistic modeling basis for the safety assessment of the operation configuration, and solves the assessment defect of related methods that only focus on the battery body and ignore the overall electrical characteristics of the energy storage power station.

[0010] In another possible implementation of the first aspect, the importance quantification result of each feature parameter is determined based on the model training process data, including: for any one of the multiple feature parameters (the first feature parameter), determining the sum of the Gini gain values ​​of the target node in each base learner from the model training process data. The target node is the node that performs feature splitting learning based on the first feature parameter. The importance quantification result of the first feature parameter is determined based on the sum of the Gini gain values ​​corresponding to each base learner and the ensemble contribution weight of each base learner.

[0011] It should be understood that this scheme, by extracting the contribution weights of each base learner and the Gini gain value of the target node, can objectively reflect the actual contribution of each feature parameter in the model decision-making process. This provides interpretable data for subsequent feature selection and weight allocation.

[0012] In another possible implementation of the first aspect, the preset importance condition includes: the importance quantization result of the feature parameter is greater than a preset quantization result threshold. And / or, the importance quantization result of the feature parameter is ranked before a preset position in the order of importance quantization results from largest to smallest.

[0013] Determine the first weight for each target feature parameter, including: normalizing the importance quantification results of each target feature parameter and determining the first weight for each target feature parameter.

[0014] It should be understood that this scheme, by setting quantitative thresholds and / or ranking positions as screening criteria, can objectively and automatically identify key features, reducing manual intervention. Furthermore, the importance of the target features is normalized to determine their weights, ensuring the objectivity, comparability, and operability of the weighting system. In another possible implementation of the first aspect, after normalizing the importance quantification results of each target feature parameter and determining the first weight of each target feature parameter, the method further includes: obtaining relative importance evaluation information for the target feature parameters and constructing a judgment matrix. The relative importance evaluation information includes the quantified values ​​of the relative importance of any two target feature parameters in the same dimension. The judgment matrix is ​​then normalized by solving for the eigenvectors to obtain the untested second weight for each target feature parameter. Based on the untested second weight for each target feature parameter, a logical consistency check is performed on the judgment matrix to determine the logical consistency check result. If the logical consistency check result is passed, the untested second weight for each target feature parameter is determined as the second weight. Based on the first weight and the second weight for each target feature parameter, the comprehensive weight for each target feature parameter is determined. Based on the first weight of each target feature parameter, the relative distance between each target feature parameter in each candidate running configuration and the desired running configuration is weighted and aggregated to determine the fit of each candidate running configuration. This includes: based on the comprehensive weight of each target feature parameter, the relative distance between each target feature parameter in each candidate running configuration and the desired running configuration is weighted and aggregated to determine the fit of each candidate running configuration.

[0015] It should be understood that this scheme integrates data-driven weights and domain knowledge weights. Subjective evaluation based on a judgment matrix is ​​introduced on top of objective weights, and consistency checks are used to ensure the logical rationality of these subjective judgments. This makes the final weight system more engineering-applicable and credible in decision-making.

[0016] In another possible implementation of the first aspect, multiple candidate operating configurations of the target energy storage power station are obtained. Based on a first weight for each target feature parameter, the relative distances between each target feature parameter in each candidate operating configuration and the desired operating configuration are weighted and aggregated to determine the fit of each candidate operating configuration. This includes: performing positive and standardized processing on the parameter values ​​of each candidate operating configuration for each target feature parameter, and constructing a decision matrix based on each processed candidate operating configuration. Each row in the decision matrix corresponds to a candidate operating configuration, and each column corresponds to a parameter value of a target feature parameter. By approximating the ideal solution and the decision matrix, the relative distance between each target feature parameter in each candidate operating configuration and the corresponding target feature parameter in the desired operating configuration is determined. The desired operating configuration includes: positive desired operating configurations and negative desired operating configurations. Positive desired operating configurations are composed of the maximum parameter value in each column of the decision matrix, and negative desired operating configurations are composed of the minimum parameter value in each column of the decision matrix. The relative distance includes: a first relative distance with the positive desired operating configuration and a second relative distance with the negative desired operating configuration. Based on the first weight of each target feature parameter, the first relative distance of each target feature parameter in each candidate running configuration is weighted and aggregated to obtain the first weighted distance from each candidate running configuration to the positive expected running configuration. Based on the first weight of each target feature parameter, the second relative distance of each target feature parameter in each candidate running configuration is weighted and aggregated to obtain the second weighted distance from each candidate running configuration to the negative expected running configuration. Based on the first weighted distance and the second weighted distance of each candidate running configuration, the fit of each candidate running configuration is determined.

[0017] It should be understood that this scheme integrates multidimensional parameters into a unified evaluation scale by positively correcting heterogeneous safety indicators and weighted aggregation based on model weights. This enables candidate operational configurations to be ranked comparablely under a quantitative system that reflects true safety preferences, thereby ensuring the rationality of the optimal results and the effectiveness of the decision.

[0018] Secondly, a processing device for configuring the operation of an energy storage power station is provided, the device comprising: The acquisition module is used to acquire model training process data for the configuration analysis model. The configuration analysis model has the ability to output the safety status results of the energy storage power station based on multiple feature parameters in the energy storage power station's operational configuration. The configuration analysis model includes multiple base learners, and each base learner includes multiple nodes. The model training process data includes: the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner during feature splitting learning based on the feature parameters during model training.

[0019] The processing module determines the importance quantification result of each feature parameter based on the model training process data. The importance quantification result characterizes the degree of influence of the corresponding feature parameter on the safety status result of the energy storage power station output by the configuration analysis model. Based on the importance quantification result, target feature parameters are selected from the feature parameters, and a first weight is determined for each target feature parameter. Target feature parameters are those whose importance quantification results satisfy preset importance conditions. The first weight is positively correlated with the degree of influence represented by the importance quantification result of the target feature parameter. Multiple candidate operating configurations of the target energy storage power station are obtained. Based on the first weight of each target feature parameter, the relative distance between each target feature parameter in each candidate operating configuration and the desired operating configuration is weighted and aggregated to determine the fit degree of each candidate operating configuration. The desired operating configuration is obtained based on the target feature parameters of multiple candidate operating configurations. The relative distance to the desired operating configuration is obtained by processing the target feature parameters of each candidate operating configuration using an approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate operating configuration and the desired operating configuration. The candidate operating configuration with the highest fit degree is determined from the candidate operating configurations and used as the target operating configuration for the target energy storage power station. The target operating configuration is used to configure the target energy storage power station. The safety status of the target energy storage power station after configuration is related to the target operating configuration.

[0020] Thirdly, a processing device for operating an energy storage power station is provided, the device comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, which includes computer instructions. When the processor executes the computer instructions, it causes the processing device for operating the energy storage power station to perform the method described in the first aspect and any possible implementation thereof.

[0021] Fourthly, a computer-readable storage medium is provided that stores computer instructions. When executed by a processor, the computer instructions are used to implement the method as described in the first aspect and any possible implementation thereof.

[0022] Fifthly, a computer program product is provided that, when run on a computer or executed by a computer's processor, implements the method described in the first aspect and any possible design thereof. The computer may be a processing device configured for operating an energy storage power station as described in the third aspect and any possible implementation thereof.

[0023] Understandably, the beneficial effects that the processing device for the operation configuration of the energy storage power station described in the second aspect, the processing equipment for the operation configuration of the energy storage power station described in the third aspect, the computer-readable storage medium described in the fourth aspect, and the computer program product described in the fifth aspect can be referred to in the beneficial effects of the first aspect and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a method for configuring the operation of an energy storage power station, as provided in an embodiment of this application. Figure 2 A flowchart illustrating a method for determining importance quantification results provided in an embodiment of this application; Figure 3 A flowchart illustrating a method for determining the comprehensive weight of target feature parameters provided in an embodiment of this application; Figure 4 A flowchart illustrating a method for determining fit according to an embodiment of this application; Figure 5 A schematic flowchart illustrating another method for configuring the operation of an energy storage power station, provided in an embodiment of this application. Figure 6 A schematic diagram of the structure of a processing device for the operation configuration of an energy storage power station provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a processing device configured for the operation of an energy storage power station, as provided in an embodiment of this application. Detailed Implementation

[0025] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0027] In the technical solutions provided in this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved are all information and data authorized by the user or fully authorized by each party. The collection, storage, use, processing, transmission, provision and disclosure of the above information and data all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0028] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0029] The operational configuration of an energy storage power station includes multiple adjustable parameters such as charging and discharging power, start-up and shutdown thresholds, and reserve capacity. While evaluation methods based on historical experience can provide an overall judgment of whether these parameter settings are "reasonable" or "safe," they cannot analyze the specific impact of each parameter on the final safety outcome. For example, when both parameters A and B exceed empirical thresholds simultaneously, existing methods can only conclude that "the configuration risk is high," but cannot determine whether parameter A has a greater impact, parameter B is the primary risk source, or whether the risk is caused by the coupling of the two. This "knowing what, but not why" evaluation model makes the causal relationship between evaluation indicators and safety status unclear.

[0030] Therefore, how to overcome the ambiguity of empirical evaluation and improve the accuracy of operational configuration decisions is an urgent problem to be solved.

[0031] In view of this, this application provides a method for processing the operation configuration of an energy storage power station. This method first analyzes the training process data of a safety assessment model to objectively quantify the importance of each operational characteristic parameter and determine its weight. Then, based on these weights, a multi-attribute decision-making method is used to calculate the degree of fit between each candidate configuration and the ideal safety state. Finally, based on the degree of fit, the optimal scheme is automatically selected from the candidate configurations, achieving objective and accurate decision-making for the operation configuration of the energy storage power station.

[0032] The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0033] The energy storage power station operation configuration processing method provided in this application embodiment can be applied to computing devices. Specifically, the computing device can be a single server or a server cluster composed of multiple servers, or a computer, or a processor or processing chip in a server or computer, etc. This application embodiment does not limit the specific device form of the computing device.

[0034] like Figure 1As shown, when the energy storage power station operation configuration processing method provided in this application embodiment is applied to the above-mentioned computing device, it specifically includes the following steps S101-S105: S101. Obtain the model training process data of the configuration analysis model.

[0035] The configuration analysis model has the ability to output the safety status results of the energy storage power station based on multiple characteristic parameters in the operation configuration of the energy storage power station. The configuration analysis model includes multiple base learners, and each base learner includes multiple nodes.

[0036] The configuration analysis model refers to a decision-making model based on ensemble machine learning algorithms, capable of predicting or classifying the safety status of an energy storage power station based on a set of input feature parameters. Base learners are the fundamental decision-making units constituting this ensemble model. Nodes are the basic units within the base learners (such as decision trees) that perform data judgment or partitioning. Multiple base learners work together through ensemble learning to improve model performance. Each node partitions the input data samples into different child nodes based on a feature parameter and its splitting condition.

[0037] Specifically, the model training process data includes: the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner when performing feature splitting learning based on the feature parameters during model training.

[0038] Ensemble contribution weights are the decision weights of each sub-model in an ensemble model, representing the magnitude of their output's influence on the final result. The Gini gain is an indicator used in decision trees to measure the improvement in classification purity when a node is split based on a certain feature; a larger value indicates a more significant contribution of that feature to the classification result at that point.

[0039] In some embodiments, the configuration analysis model may be trained locally on the computing device based on a historical runtime configuration dataset, or it may be downloaded by the computing device from other devices (e.g., a data center). This application embodiment does not limit the method of obtaining the configuration analysis model.

[0040] One possible implementation involves the computing device using an Adaptive Boosting algorithm (AdaBoost) to train a random forest model, thereby obtaining the configuration analysis model and acquiring data from the model training process. In this case, the base learner is specifically a base classifier. Specifically, the process may include: 1. Constructing training data. The training data includes the historical operating configurations of multiple sample energy storage power stations. Each historical operating configuration includes multiple feature parameters and the actual safety status label after being applied to the corresponding sample energy storage power station. The structure of this training data is shown in Equation (1):

[0041] in, The feature parameter vector configured for the historical operation of the i-th energy storage power station includes the parameter values ​​of multiple feature parameters. This is the actual safety status label for the i-th energy storage power station. A value of -1 indicates that the energy storage power station is in a safe state, and a value of +1 indicates that the energy storage power station is in a safe state.

[0042] 2. Initialize the weights of the model training data. Set the training rounds to M and initialize the weights of each training sample in the training data. As shown in equation (2):

[0043] in, The initial weights are for the nth feature parameter.

[0044] 3. Iterative training. For the m-th iteration ( The computing device first uses the current sample weight distribution ( Weighted sampling or weighted learning is performed to train the m-th base classifier. Next, for the evaluation Performance, computation The weighted classification error rate under the current weight distribution.

[0045] For example, when iterating through all N training samples, for each sample, the first step is to determine the result using an indicator function. The prediction is not equal to the true label of the sample. If the prediction is wrong, the indicator function outputs 1, otherwise it outputs 0. Then, this 0 / 1 judgment result of "whether it is wrong" is multiplied by the weight of the sample in the current iteration. Finally, the weighted error values ​​of all samples are added together, and the sum is the weighted error rate. The calculation process is shown in equation (3):

[0046] in, This represents the weight of the i-th training sample in the m-th iteration. This represents the predicted safety state of the m-th base classifier for the i-th training sample. This is an indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise.

[0047] Then, based on the error rate of the base classifier, the ensemble contribution weight of that base classifier in the configuration analysis model is determined. This weight coefficient... The calculation process is shown in equation (4):

[0048] Where, in the formula, This represents the ensemble contribution weight of the m-th base classifier. This represents the weighted classification error rate of the m-th base classifier. It is calculated by adjusting the classifier advantage ratio. Take the natural logarithm and multiply it by the coefficient 'a' to calculate the error rate. It is mapped to a weight value that represents the confidence level of the classifier's decision.

[0049] When the weighted error rate of the base classifier When the odds ratio is less than 0.5 (i.e., better than random guessing), the odds ratio is greater than 1, its logarithm is positive, and... The smaller, The larger the value, the greater the contribution of the base classifier to the final ensemble model. The coefficient 'a' is a constant introduced in the theoretical derivation of the algorithm to simplify the subsequent weight update formula.

[0050] Next, to ensure that the base classifiers generated in the next iteration are more focused on correcting the classification errors of the current model, the computing device needs to update the weights of each training sample. The update principle is based on the prediction results of the current base classifiers and their weights. This increases the weight of misclassified samples while decreasing the weight of correctly classified samples, thus creating a new sample weight distribution. Used for subsequent training. The update process of sample weights is shown in equation (5):

[0051] In equation (5), This represents the weight of the i-th training sample in the (m+1)-th iteration after the update. It is a normalization factor, and its calculation process is shown in equation (6):

[0052] Normalization factor Its purpose is to ensure that all weights after the update The sum of these terms is 1, thus forming an effective probability distribution. In equation (5), the product term... This constitutes a judgment on the correctness of sample classification. When the base classifier When the prediction is correct, and When the signs are the same, their product is +1, and the exponent term becomes... , It is a value less than 1, so the weight of this sample will be reduced. When the prediction is wrong, its product is -1, and the exponent becomes... , It is a value greater than 1, therefore the weight of this sample will be amplified. The degree of weight amplification is determined by the ensemble contribution weights of the base classifiers in this round. The better the control and performance of the classifier, the better. The larger the error (the more significant the error), the greater the weight of the corresponding sample will be, ensuring that the model can focus on learning these samples that are difficult to classify correctly in subsequent iterations.

[0053] After completing all m rounds of iterative training, the computing device will combine all base classifiers. According to its corresponding integration contribution weight By performing linear combinations, the final configuration analysis model (i.e., a strong classifier) ​​is constructed. Its decision-making process is shown in equation (7):

[0054] Based on this, and using the trained configuration analysis model and its process data, the computing device performs processing and analysis on the Gini gain values ​​in the model training process data.

[0055] First, calculate the Gini index of each node t in each base classifier (decision tree) before and after the split. Gini index The impurity of the sample labels contained in node t is measured and its calculation process is shown in equation (8):

[0056] In the formula, C is the number of categories in the decision tree classification (e.g., C=2). The Gini index represents the proportion of samples of class i in node t. The higher the purity of a node, the higher the Gini index. The lower the value, the better. When node t is split based on characteristic parameter A (i.e., any one of the characteristic parameters), the Gini coefficient after the split is... The weighted average of the Gini indices of the two child nodes generated by the split is calculated as shown in equation (9):

[0057] In the formula, N represents the total number of samples contained in node t. The left child node after splitting The number of samples, The right child node after splitting The number of samples. and These are the Gini indices for the left and right child nodes, respectively.

[0058] Next, the Gini gain contributed by each feature parameter in each base classifier is calculated. Node t is defined as the Gini gain contributed by feature parameter A in this split, based on the difference in the Gini index before and after the feature parameter split. The calculation process is shown in equation (10):

[0059] Gini gain value The degree of impurity reduction caused by this splitting action was quantified. The larger the value, the more significant the effect of using feature A to split at this node on improving classification purity.

[0060] Ultimately, the computing device not only acquired the configuration analysis model, but also obtained and recorded the complete model training process data.

[0061] In other embodiments, the computing device may also employ a gradient boosting decision tree algorithm. In this algorithm, each decision tree (base learner) sequentially fits the prediction residuals of the previous stage, and its ensemble contribution weight can be set by a preset learning rate or other means to reflect the contribution of each tree to the overall model. Simultaneously, when a node splits, each tree also calculates a corresponding index value based on preset splitting criteria (such as Gini gain, information gain, or mean squared error reduction) to measure the contribution of the corresponding feature parameters to constructing a better splitting rule.

[0062] S102. Based on the model training process data, determine the importance quantification result of each feature parameter.

[0063] Among them, the importance quantification results are used to characterize the degree of influence of the corresponding feature parameters on the safety status results of the energy storage power station output by the configuration analysis model.

[0064] Specifically, the output of the configuration analysis model is determined by a weighted average of the contributions of each base learner based on its ensemble contribution. Each base learner's decision depends on node splits, and the Gini gain of a node directly quantifies the contribution of the used features to that split. Therefore, the impact of feature parameters on the model output can be determined by their total Gini gain (local contribution) within each base learner, weighted by the learner's ensemble contribution.

[0065] In some embodiments, when determining the importance quantification result of each feature parameter based on model training process data, the computing device may employ a weighted aggregation computing framework.

[0066] Specifically, within each base learner, the computing device can accumulate the Gini gain value contributed by a specific feature on all relevant nodes to obtain the local importance of the feature in this single model. Then, considering that different base learners have different decision weights in the final ensemble model, the above local importance is weighted and summed using the ensemble contribution weights of each base learner, thereby fusing the evaluations from different confidence models to obtain the quantitative result of the importance of the feature.

[0067] The specific process of this embodiment can be referred to in the following text. Figure 2 The details of the matter will not be elaborated here.

[0068] In some embodiments, when determining the importance quantification of each feature parameter based on model training process data, the computing device may employ a statistical aggregation framework based on a combination of frequency and gain. This framework comprehensively evaluates the feature importance by analyzing the frequency of feature usage across the entire model and the average effectiveness of each usage.

[0069] One possible implementation involves the computing device first traversing all base learners and their nodes, counting the total number of times each feature parameter is selected as a splitting feature, denoted as the frequency index. Next, for each feature parameter, the computing device extracts its corresponding Gini gain value across all used nodes and calculates the arithmetic mean or sum of these Gini gain values, denoted as the average gain index. Finally, the computing device multiplies the frequency index by the average gain index; the resulting product is the importance quantification result for that feature parameter.

[0070] S103. Based on the importance quantification results, select target feature parameters from the feature parameters and determine the first weight of each target feature parameter.

[0071] The target feature parameter is the feature parameter whose importance quantification result meets the preset importance condition. The first weight is positively correlated with the degree of influence represented by the importance quantification result of the target feature parameter.

[0072] Specifically, to eliminate interference from noise features, the computing device first filters out key target feature parameters based on the importance quantification results, eliminating features with low contribution. Then, based on the importance of the target features, a first weight proportional to their influence is determined through methods such as normalization, thereby focusing on key influencing factors.

[0073] For example, the importance quantification result can be a normalized value between 0 and 1, a non-negative original weighted score, or a score after specific scaling processing, etc. The embodiments of this application do not limit the specific numerical form of the importance quantification result.

[0074] In some embodiments, the preset importance condition may include the importance quantification result of the feature parameter being greater than a preset quantification result threshold.

[0075] The preset threshold for the quantification result can be a fixed value set manually based on experience or understanding of the model, or it can be a relative value dynamically calculated based on the statistical distribution of the quantification results of the importance of all features (such as the mean, median, upper quartile, etc.) to achieve adaptive filtering.

[0076] One possible implementation is that the computing device stores a preset fixed threshold (e.g., 0.05) or a dynamically calculated threshold (e.g., half the average importance of all features) in a configuration file. During filtering, the computing device reads this threshold and compares it with the importance quantification result of each feature one by one, filtering out features that are greater than the threshold.

[0077] For example, the computing device can set a threshold as the average of the quantified importance results of all features. Assuming the calculated average importance of all features is 0.1, then the threshold is 0.1. If the importance result of the feature "grounding resistance value parameter" is 0.15, which is higher than the average of 0.1, then it is selected as the target feature parameter; if the importance result of the feature "relay protection device model parameter" is 0.08, which is lower than the average of 0.1, then it is not selected as the target feature parameter.

[0078] In other embodiments, the preset importance condition may include the importance quantification result of the feature parameter being ranked before a preset position in a sorting process based on the importance quantification result from largest to smallest.

[0079] Similarly, the preset ranking can be a fixed value set manually based on the number of features to be retained (such as the top 5), or it can be a cutoff ranking dynamically determined based on the cumulative importance contribution ratio (such as which features contribute to the top 80% importance).

[0080] One possible implementation is that the computing device sorts all features in descending order according to their importance quantification results. If a fixed rank K is preset, the computing device directly selects the top K features from the sorted list. If a cumulative contribution ratio P is preset (e.g., 80%), the computing device sequentially adds the importance of each feature starting from the first-ranked feature until the cumulative sum reaches or exceeds P% of the total importance. These features participating in the accumulation are then identified as the target feature parameters.

[0081] For example, if there are 10 features, and after sorting them by importance, the cumulative importance of the top 3 features reaches 85% of the total importance. If the preset cumulative contribution ratio P is 80%, then the computing device selects these top 3 features as the target feature parameters.

[0082] In other embodiments, the preset importance condition may also be a logical combination of the above-mentioned quantization result threshold condition and the ranking condition.

[0083] For example, the computing device can be configured with the following conditions: the importance quantification result must be greater than a threshold of 0.03 and the ranking must be within the top 10. Assuming feature A has an importance result of 0.05 and ranks 8th among all features, it satisfies both sub-conditions and is selected as the target feature parameter. Assuming feature B has an importance result of 0.04 and ranks 12th, it satisfies the threshold condition but not the ranking condition, and therefore is not selected. Assuming feature C has an importance result of 0.02 and ranks 5th, it satisfies the ranking condition but not the threshold condition, and is also not selected.

[0084] In some embodiments, the computing device may normalize the importance quantification results of each target feature parameter to determine a first weight for each target feature parameter.

[0085] Specifically, the computing device first sums the importance quantification results of all target feature parameters to obtain a total importance value. Then, the importance quantification result of each target feature parameter is divided by this total importance value; the quotient is the first weight of that feature parameter. This process can be expressed as:

[0086] in, This represents the first weight of the j-th target feature parameter. This represents the importance quantification result of the j-th target feature parameter. This represents the total number of target feature parameters. This represents the sum of the importance quantification results of all target feature parameters. Through this normalization process, the sum of the first weights of each target feature parameter is 1, and the weights... Its corresponding original importance Proportional.

[0087] For example, if the selected target features are X, Y, and Z, and their importance quantification results are 0.8, 0.5, and 0.2 respectively, then the total importance value is 1.5. After normalization, the first weight of feature X is 0.8 / 1.5≈0.533, the first weight of feature Y is 0.5 / 1.5≈0.333, and the first weight of feature Z is 0.2 / 1.5≈0.133.

[0088] It should be understood that this scheme, by extracting the contribution weights of each base learner and the Gini gain value of the target node, can objectively reflect the actual contribution of each feature parameter in the model decision-making process. This provides interpretable data for subsequent feature selection and weight allocation.

[0089] S104. Obtain multiple candidate operating configurations for the target energy storage power station. Based on the first weight of each target feature parameter, perform weighted aggregation on the relative distance between each target feature parameter and the desired operating configuration in each candidate operating configuration to determine the fit of each candidate operating configuration.

[0090] The desired running configuration is obtained based on the target feature parameters of multiple candidate running configurations. The relative distance to the desired running configuration is obtained by processing the target feature parameters of each candidate running configuration using an approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate running configuration and the desired running configuration.

[0091] Candidate operating configurations refer to a set of different characteristic parameters that can be selected for a target energy storage power station. Each candidate operating configuration sets a specific set of parameter values ​​for each target characteristic parameter. The approximation ideal solution method compares the performance of each candidate scheme on different target characteristic parameters, constructs a theoretically optimal scheme (called the positive ideal solution) and a theoretically worst scheme (called the negative ideal solution), and calculates the distance between each candidate scheme and these two ideal schemes to evaluate its merits.

[0092] Specifically, by using the first weight of the target feature parameters, the relative distances between each candidate operating configuration and the desired operating configuration are weighted and aggregated across each feature dimension. This ensures that the evaluation results objectively reflect the true impact of each feature parameter on the security status. This method allows high-weighted feature parameters to dominate the distance calculation, thus enabling the final fit index to more accurately characterize the degree to which each candidate operating configuration approaches the ideal state in key security dimensions.

[0093] In some embodiments, candidate runtime configurations can be pre-screened using a configuration analysis model. The computing device can input the feature parameter vector of each candidate runtime configuration into a trained configuration analysis model to obtain the predicted security status output by the model. Based on the prediction results (e.g., security score or classification label), the computing device filters out candidate runtime configurations with security statuses below a preset security threshold.

[0094] In some embodiments, the computing device may employ an extreme value construction method to obtain the desired operating configuration based on the target feature parameters of multiple candidate operating configurations. Specifically, the computing device traverses all candidate operating configurations, and for each target feature parameter, selects the maximum value (benefit-oriented) or minimum value (cost-oriented) of all configurations on that feature to form a positive desired operating configuration; and selects the minimum value (benefit-oriented) or maximum value (cost-oriented) to form a negative desired operating configuration.

[0095] In some embodiments, when a computing device uses a first weight to perform weighted aggregation to determine the fit, the specific execution process can be divided into several steps: data preprocessing, constructing a decision matrix, determining the desired configuration, calculating the relative distance and weighted aggregation.

[0096] The specific process of this embodiment can be referred to below. Figure 4 The details of the matter will not be elaborated here.

[0097] In some embodiments, the computing device may also incorporate additional weight information to form a comprehensive weight for fit calculation before performing weighted aggregation, thereby incorporating other rational considerations on a data-driven basis.

[0098] One possible implementation is that the computing device can subjectively analyze the target feature parameters based on domain knowledge or expert experience to assess the relative importance of the target feature parameters, in order to supplement or correct the first weight obtained by pure data-driven analysis.

[0099] The specific process of this embodiment can be referred to below. Figure 3 The details of the matter will not be elaborated here.

[0100] S105. Determine the candidate operating configuration with the highest fit from the candidate operating configurations, and use it as the target operating configuration for the target energy storage power station.

[0101] The target operating configuration is used to configure the target energy storage power station, and the safety state of the configured target energy storage power station is related to the target operating configuration. This target operating configuration will be used as a control parameter in the actual operation of the power station, directly constraining and influencing its subsequent actual safety level. This method selects the configuration with the highest fit through quantitative evaluation, that is, sets a set of operating parameters for the power station that best approximates the desired safety state in the feature space and can optimally balance the various safety dimensions, aiming to guide the system to achieve a better overall safety state in actual operation.

[0102] Specifically, a higher fit value indicates a greater degree of fit between the configuration and the expected operating configuration, and correspondingly, its security performance is closer to the theoretically expected state.

[0103] In some embodiments, if multiple candidate running configurations have the same fit and are all at the maximum value, the computing device may use auxiliary decision rules to make the final decision.

[0104] One possible implementation involves the computing device comparing the performance of these parallel configurations on the most important feature dimension. Specifically, the computing device first identifies the target feature parameter with the largest first weight. Then, it calculates the relative distance (or directly compares their standardized parameter values) between each of these parallel configurations and the expected operating configuration on that feature. Finally, the configuration that performs best on that feature (i.e., has the smallest relative distance or the largest benefit parameter value) is selected as the target operating configuration.

[0105] For example, suppose that among the target characteristic parameters, "peak inter-pole fault current" has the largest first weight. The fit of both Config_M and Config_N is 0.95. The computing device compares the relative distances of both to the desired configuration on the feature of "peak inter-pole fault current," finding that Config_M's distance is 0.1 and Config_N's distance is 0.15. Since Config_M is closer to the ideal value on the most important feature, the computing device ultimately selects Config_M as the target operating configuration.

[0106] This application provides a method for processing the operation configuration of an energy storage power station. First, by analyzing the Gini gain value generated by feature splitting at each node during model training, and the contribution weight of each base learner in the ensemble model, the method utilizes the feature selection and combination capabilities automatically performed by the model during training to improve output accuracy. This allows for reverse reasoning of the implicit decision-making logic within the model to quantify the importance of each feature parameter to the final safety state result. Subsequently, the feature parameters are screened for importance, and the normalized weights of the target feature parameters are determined. Compared to related methods, this application can objectively screen and weight feature parameters (evaluation indicators) based on the inherent objective laws of the data learned during model training, reducing the disconnect between the evaluation results and the noise indicators introduced by relying on historical experience. This constructs a feature evaluation system that more accurately reflects the actual safety state of the energy storage power station. Furthermore, when evaluating and deciding on candidate operation configurations for the energy storage power station, the relative distance between each candidate operation configuration and the desired operation configuration (defined by positive and negative ideal solutions) is calculated using an approximation ideal solution method, and the weights of the aforementioned target feature parameters are introduced to weighted aggregate these distances. This mechanism ensures that key characteristic parameters significantly impacting the safety status of energy storage power stations are objectively and reasonably emphasized during the evaluation of candidate operating configurations. This allows the final fitted score to more accurately assess the true effectiveness of each candidate operating configuration, thus mitigating the risk of discrepancies between evaluation results and actual effects in related technologies. Finally, by selecting the configuration with the highest fitted score as the target operating configuration, a complete decision-making loop is achieved, from objective data pattern mining to key parameter weight determination, and then to accurate evaluation and automatic optimization of multiple schemes. This improves the accuracy of energy storage power station operation decisions. It provides a direct and reliable basis for the daily operation and safety strategy formulation of energy storage power stations, enhancing their safety performance.

[0107] The following describes the feature parameters involved in the embodiments of this application: In some embodiments, the characteristic parameters include at least one of the following dimensions: DC-side battery pack grounding characteristic dimension, power conversion characteristic dimension, relay protection characteristic dimension, and fault characteristic dimension. The DC-side battery pack grounding characteristic dimension includes: DC battery pack connection method parameters and grounding resistance value parameters. The power conversion characteristic dimension includes: converter operating condition parameters, converter performance parameters, and converter safety protection function parameters. The relay protection characteristic dimension includes: relay protection action criterion type parameters, relay protection area range parameters, relay protection device model parameters, and relay protection function parameters. The fault characteristic dimension includes: inter-pole fault current peak parameters, polar fault current peak parameters, positive and negative DC bus voltage peak parameters, DC energy storage capacitor fault voltage parameters, AC side three-phase current offset parameters, polar fault current response time parameters, undervoltage protection response time parameters, and converter blocking response time parameters.

[0108] The grounding characteristic dimension of the DC-side battery pack includes parameters describing the electrical connection characteristics between the DC-side battery pack and the ground of the energy storage power station, such as DC battery pack grounding method parameters and grounding resistance value parameters. The grounding method parameters can indicate whether ungrounded, resistor-grounded, or inductively grounded modes are used; the grounding resistance value parameters quantify the magnitude of the resistance in the grounding loop that limits the fault current.

[0109] The power conversion characteristic dimension includes parameters used to describe the operating status and capabilities of the power conversion system (PCS) in the energy storage power station that realizes AC / DC power conversion. These parameters include converter operating condition parameters, converter performance parameters, and converter safety protection function parameters. Converter operating condition parameters may include the allowable voltage and frequency operating range; performance parameters may include conversion efficiency and overload capacity; and safety protection function parameters may refer to whether the system has the protection functions required by grid connection standards (such as overvoltage, undervoltage, overfrequency, and underfrequency protection).

[0110] The relay protection characteristic dimension includes parameters describing the configuration and characteristics of the internal electrical protection system of the energy storage power station, such as relay protection action criterion type parameters, relay protection area parameters, relay protection device model parameters, and relay protection function parameters. Action criterion type parameters can distinguish between current protection, voltage protection, differential protection, etc.; area parameters define the range of electrical equipment covered by the protection (such as battery clusters, DC buses, transformers, etc.); device model and function parameters specifically describe the hardware capabilities of the protection equipment and the protection logic implemented.

[0111] The fault characteristic dimension includes parameters describing the electrical quantity and time response characteristics of the energy storage power station under simulated or historical fault events. These parameters include peak inter-pole fault current parameters, peak polar fault current parameters, peak positive and negative DC bus voltage parameters, DC energy storage capacitor fault voltage parameters, AC side three-phase current offset parameters, polar fault current response time parameters, undervoltage protection response time parameters, and converter blocking response time parameters. These parameters quantify the electrical stress intensity (such as current and voltage peaks) and the operating speed of critical system protection components during a fault, directly reflecting the power station's fault tolerance and clearing capabilities.

[0112] Furthermore, assessments of operational configurations in related technologies often focus on the safety of the battery itself, lacking a structured, multi-dimensional, quantitative description method from the perspective of system electrical safety. This makes it difficult to systematically analyze and objectively compare the electrical safety impact of operational configurations. In contrast, the feature parameters in this embodiment include electrical characteristic dimensions affecting the overall safety of the energy storage power station. By concretizing the abstract operational configuration into a quantifiable parameter system, it provides a comprehensive and engineering-practical modeling foundation for the safety assessment of operational configurations, overcoming the assessment deficiency of related methods that only focus on the battery itself while neglecting the overall electrical characteristics of the energy storage power station.

[0113] The following describes the process of determining the quantification results of the importance of each feature parameter.

[0114] In some embodiments, as described in the example S102 above, when determining the importance quantification result of each feature parameter based on model training process data, the computing device may employ weighted aggregation. This process is as follows... Figure 2 As shown, S102 specifically includes the following steps S201-S202: S201. For any one of the multiple feature parameters, determine the sum of the Gini gain values ​​of the target node in each base learner from the model training process data.

[0115] The target node is the node that learns feature splitting based on the first feature parameter.

[0116] Specifically, the Gini gain reflects the local contribution of a feature when splitting at a single node. To quantify the overall contribution of the first feature parameter within a base learner, the computing device needs to traverse all nodes of the learner, select the target nodes for splitting based on that feature, and sum their Gini gains. This sum represents the total contribution of that feature in the construction of the decision rules for this learner.

[0117] In some embodiments, when determining the sum of the Gini gain values ​​of the target nodes in each base learner, the computing device may employ a node traversal and conditional accumulation method.

[0118] One possible implementation involves the computing device reading the structure data of the base learner, which records the feature identifier used by each node and its corresponding Gini gain value. The computing device iterates through all nodes, and if a node's feature identifier matches a first feature parameter, its Gini gain value is accumulated. The accumulated sum after the iteration is completed represents the local contribution of that feature within this learner.

[0119] For example, in a certain base learner, the feature "peak inter-electrode fault current" is used for splitting at three nodes, with Gini gains of 0.05, 0.08, and 0.03, respectively. The sum of these gains yields a local contribution of 0.16.

[0120] S202. Based on the sum of the Gini gain values ​​corresponding to each base learner and the ensemble contribution weight of each base learner, determine the importance quantification result of the first feature parameter.

[0121] One possible implementation is that the computing device determines the importance quantification result through global normalization. Specifically, the ratio of the total Gini gain of the first feature parameter in all base learners to the total Gini gain of all features is calculated, and this ratio is the importance quantification result. The calculation process is shown in Equation (12).

[0122]

[0123] in, This represents the quantification result (i.e., the degree of importance) of the j-th feature parameter. Let M represent the sum of the Gini gain values ​​contributed by the j-th feature parameter in the m-th base learner, where M is the total number of base learners and N is the total number of feature parameters.

[0124] Another possible implementation is that the computing device can first independently normalize the feature contributions within each base learner, and then perform integrated weighting.

[0125] Specifically, for the m-th base learner, the computing device first normalizes the sum of the Gini gain values ​​of its internal feature parameters to obtain a set of local importance distributions within that learner. Then, the computing device uses the ensemble contribution weights of the base learners to weight these local importance values. Finally, the computing device sums the weighted local importance values ​​of all base learners and performs global normalization on the summation result again to obtain the final importance quantification result.

[0126] It should be understood that this scheme, by extracting the contribution weights of each base learner and the Gini gain value of the target node, can objectively reflect the actual contribution of each feature parameter in the model decision-making process. This provides interpretable data for subsequent feature selection and weight allocation.

[0127] In some embodiments, in addition to determining the target feature parameters and their (first) weights based on data obtained during model training, subjective analysis results from domain experts can be incorporated to assign weights to the target feature parameters based on another evaluation dimension. This process is as follows: Figure 3 As shown, after S103, the method further includes steps S301-S306: S301. Obtain the relative importance assessment information for the target feature parameters and construct the judgment matrix.

[0128] The relative importance assessment information includes the quantitative values ​​of the relative importance of any two target feature parameters under the same dimension.

[0129] Specifically, the computing device acquires pairwise comparison evaluation information for the selected target feature parameters. This information quantifies the relative importance between any two target feature parameters in numerical form (e.g., using a 1-9 scale, where 1 represents equal importance and 9 represents absolute importance). Based on these pairwise comparison results, the computing device can construct a judgment matrix, where the rows and columns of the matrix correspond to the target feature parameters, as shown in Equation (13):

[0130] Where A represents the judgment matrix, which is an n×n square matrix, and n is the number of target feature parameters. Elements (like , ) represents the importance comparison scale value of the i-th target feature parameter relative to the j-th target feature parameter.

[0131] Matrix A must satisfy the following mathematical constraints: the values ​​of two elements symmetric about the main diagonal are reciprocals of each other, the values ​​of the diagonal elements are all positive, and the values ​​of the off-diagonal elements are positive, as shown in equation (14):

[0132] In some embodiments, the values ​​of the elements of the determination matrix can be determined using a 1-9 scale, as shown in Table 1: Table 1

[0133] It should be understood that the 1-9 scaling method shown in Table 1 is only an exemplary quantification rule. In practical applications, the values ​​of the matrix elements can follow other predefined scaling systems, such as the 0-2 scaling method, the exponential scaling method, or a custom discrete scaling set. This application does not limit the specific scaling type used, as long as it can quantify the relative influence between the two target feature parameters.

[0134] S302. Solve the normalized eigenvectors of the judgment matrix to obtain the second weight to be verified for each target feature parameter.

[0135] One possible implementation is that the computing device can use the square root method to solve the problem. This method calculates and normalizes the geometric mean of the elements in each row of the judgment matrix to approximately obtain the eigenvector corresponding to the largest eigenvalue of the matrix. This vector is the set of test weights for each target feature parameter.

[0136] Specifically, first, calculate the product of all elements in each row of the judgment matrix, then find its nth root to obtain the intermediate value. The calculation process is shown in equation (15):

[0137] In the formula, n is the number of target feature parameters. To quantify the importance of the i-th feature relative to the j-th feature in the matrix, This represents the element in each row (i represents the row, j represents the column). ( ) Perform a series of multiplications.

[0138] Then, for by The resulting vector is normalized so that the sum of its elements is 1, thus obtaining the final second weight to be verified. The normalization process is shown in equation (16):

[0139] For example, for a 3rd order (n=3n=3) judgment matrix, the computing device first calculates the geometric mean of each row according to equation (15): =2.0、 =0.5、 =0.25, then The value is 2.75. Subsequently, the computing device normalizes the values ​​according to equation (16) to obtain the second weights to be verified for each target feature parameter: =2.0 / 2.75≈0.727, =0.5 / 2.75≈0.182, =0.25 / 2.75≈0.091.

[0140] S303. Based on the second weight to be verified for each target feature parameter, perform a logical consistency check on the judgment matrix and determine the logical consistency check result of the judgment matrix.

[0141] The logical consistency test utilizes the solved second weight to quantify the degree of inconsistency between it and the original judgment matrix. Theoretically, if the judgment matrix satisfies the mathematical condition of complete consistency (i.e., any element in the matrix satisfies transitivity), then... Then its largest eigenvalue The eigenvectors are equal to the matrix order n, and the corresponding eigenvectors are the accurate weight vectors. However, in practical applications, expert subjective judgments are unlikely to achieve perfect mathematical consistency, leading to logical inconsistencies. Therefore, it is necessary to calculate the largest eigenvalue. The deviation from n is used to objectively measure the magnitude of the inherent contradiction (i.e., inconsistency) in the expert judgment, and to determine whether the inconsistency is acceptable based on a statistical threshold.

[0142] In some embodiments, the computing device may perform the verification by calculating a consistency ratio.

[0143] For example, the process specifically includes: 1. The computing device determines the product vector AW of the judgment matrix A and the weight vector (set) W, as shown in equation (17):

[0144] The values ​​of each element in AW are: .

[0145] 2. Determine the cumulative sum of each row to form a vector. , Each element in the vector It is the sum of all elements in the i-th row of matrix AW, as shown in equation (18):

[0146] 3. Calculate the largest eigenvalue of the judgment matrix using the product vector AW and the weight vector W. The approximate value is calculated using the following formula:

[0147] For a matrix A and its eigenvector W, AW = λW, i.e., a new vector... Each component It should be equal to the corresponding component of the original vector. The ratio is λ times (where λ is the eigenvalue). Therefore, for each i, the ratio is... / Theoretically, all ratios should be equal to the eigenvalue λ. However, since actual matrices are not perfectly identical, these ratios are not exactly equal; therefore, their arithmetic mean is taken as the largest eigenvalue. This is an approximate estimate. The closer this value is to the matrix order n, the better the consistency of the matrix.

[0148] 4. The computing device calculates based on the maximum eigenvalue. Calculate the consistency index Next, find the corresponding average random consistency index. (This value is a constant related to the matrix order). Finally, calculate the consistency ratio. .like If the value is less than 0.10, the computing device determines that the logical consistency test result is passed; otherwise, it is determined to fail.

[0149] For example, where, The average random consistency test index for each order of judgment matrix is ​​shown in Table 2. Table 2

[0150] For example, for a 3rd order judgment matrix, its... =0.028, from the table, the 3rd order matrix is... =0.58, then =0.028 / 0.58≈0.048<0.10, then the logical consistency test result of the judgment matrix is ​​passed.

[0151] S304. If the logical consistency test result is passed, the second weight to be tested for each target feature parameter is determined as the second weight.

[0152] Specifically, the purpose of the logical consistency test is to verify whether the internal consistency of the expert judgment is acceptable. If the test passes (e.g., consistency ratio CR < 0.10), it indicates that the second weights to be tested, calculated based on the current judgment matrix and corresponding to each target feature parameter, are logically consistent and credible, and can be used as effective weights reflecting the expert's subjective evaluation or domain knowledge preference for these key feature parameters.

[0153] In some embodiments, if the logical consistency check fails, the computing device may trigger a weight correction or information re-acquisition process.

[0154] One possible implementation is that when the verification fails, the computing device outputs a prompt message (such as "Expert consensus has not met the requirements, please re-evaluate") and returns the process to S301, so that domain experts can review and adjust the pairwise comparison evaluation information and reconstruct the judgment matrix. Alternatively, the computing device can also use a preset matrix correction algorithm (such as the minimum adjustment method) to automatically fine-tune the original judgment matrix to meet the consistency requirements, and recalculate the second weight based on the corrected matrix.

[0155] S305. Based on the first weight and the second weight of each target feature parameter, determine the comprehensive weight of each target feature parameter.

[0156] Specifically, the first weight originates from the data-driven model training process, objectively reflecting the degree of influence of features on historical security status; the second weight originates from expert experience or domain knowledge, subjectively reflecting the relative importance of features in the current decision-making scenario. Integrating these two types of weight information allows us to obtain target feature parameter weight coefficients that simultaneously consider both objective data patterns and subjective domain experience.

[0157] One possible implementation is that the computing device assigns a preset fusion coefficient (e.g., an objective coefficient β and a subjective coefficient 1-β, where 0 ≤ β ≤ 1) to the first weight and the second weight respectively. For the i-th target feature parameter, the computing device assigns its first weight... With the second weight Computing devices can be configured according to The comprehensive weights of the i-th target feature parameters are obtained by fusion. .

[0158] For example, if we believe the model's learning results are more reliable, we can set β=0.7. For a certain target feature parameter, its first weight is 0.4 and its second weight is 0.6, then its combined weight is 0.7×0.4+0.3×0.6=0.46.

[0159] Another possible implementation is that, for each target feature parameter, the computing device calculates the geometric mean of its first weight and second weight, and then normalizes this mean (so that the sum of the combined weights of all features is 1) to obtain the combined weight. The calculation formula can be expressed as: For example, for the i-th target feature parameter, its comprehensive weight It can be calculated using the following formula:

[0160] in, and Let represent the first weight and the second weight of the i-th target feature parameter, respectively, and NE be the total number of target feature parameters. This formula represents the summation of the geometric mean of all target feature parameters to achieve normalization, ensuring that the sum of the comprehensive weights is 1. This formula achieves fusion through geometric mean and is sensitive to extremely small weight values ​​from either source, emphasizing the consistency between the two weighting opinions.

[0161] For example, suppose there are three target feature parameters, with their first and second weights as follows: Target feature parameter 1: =0.6, =0.2; Target feature parameter 2: =0.3, =0.5; Target feature parameter 3: =0.1, =0.3. First, calculate the geometric mean of each feature: 0.6×0.2≈0.346, 0.3×0.5≈0.387, 0.1×0.3≈0.173. The sum of the geometric means is 0.346+0.387+0.173=0.906, then the comprehensive weight of each target feature parameter is: =0.346 / 0.906≈0.382, =0.387 / 0.906≈0.427, 0.173 / 0.906≈0.191.

[0162] In this case, S104 can be implemented as follows: S306. Based on the comprehensive weight of each target feature parameter, the relative distance between each target feature parameter and the desired running configuration in each candidate running configuration is weighted and aggregated to determine the fit of each candidate running configuration.

[0163] The process of this embodiment can be referred to the description of S104 above and the following text. Figure 4 And its detailed description.

[0164] It should be understood that this scheme integrates data-driven weights and domain knowledge weights. Subjective evaluation based on a judgment matrix is ​​introduced on top of objective weights, and consistency checks are used to ensure the logical rationality of these subjective judgments. This makes the final weight system more engineering-applicable and credible in decision-making.

[0165] The following section details the process for determining the fit of each candidate runtime configuration.

[0166] In some embodiments, when a computing device determines the fit by performing weighted aggregation using a first weight, the specific execution process can be divided into several steps: data preprocessing, constructing a decision matrix, determining the desired configuration, and calculating relative distance and weighted aggregation. This process is as follows: Figure 4 As shown, S104 specifically includes the following steps S401-S405: S401. For each candidate running configuration, perform positive transformation and standardization on the parameter values ​​of each target feature parameter, and construct a decision matrix based on each candidate running configuration after processing.

[0167] In the decision matrix, each row corresponds to a candidate running configuration, and each column corresponds to the parameter value of a target feature parameter.

[0168] Each target characteristic parameter has a different physical meaning and dimension, and their values ​​may affect the safety status in different directions (benefit-oriented or cost-oriented). Therefore, computing devices can eliminate the dimensional differences in the target characteristic parameter values ​​through standardization, and, based on positive processing, ensure that the evaluation direction of all parameters is consistent (i.e., the larger the value, the better the safety status).

[0169] In some embodiments, the computing device can classify the parameter values ​​of the target feature parameters into extremely small, intermediate, and extremely large types according to their influence on the safety status of the energy storage power station, and perform positive processing on the parameter values ​​of different types.

[0170] For example, the "fault response time parameter" (such as polar fault current response time and undervoltage protection response time) among the aforementioned characteristic parameters is usually very small; that is, the smaller the parameter value, the faster the fault handling and the better the safety status. The "grounding resistance value parameter" may be intermediate under certain safety standards, meaning the resistance value needs to be stable within a specific range (neither too large, leading to insufficient insulation, nor too small, leading to excessive fault current); the closer to the center value of this ideal range, the better the safety. Conversely, "certain performance rating parameters" or "safety margin parameters" are usually very large; that is, the larger the parameter value, the higher the performance or margin, and the better the safety status.

[0171] Among them, the feature parameters of extremely large targets do not need to be forwarded because the larger the value, the better the safety status. Their original parameter values ​​can be directly used in subsequent standardization steps.

[0172] One possible implementation involves the computing device performing forward processing using a difference transformation method for extremely small target feature parameters. Specifically, the computing device first obtains the maximum value of the feature parameter across all candidate runtime configurations. For the i-th candidate runtime configuration, the original parameter value for that feature parameter is then calculated. Calculate its positiveized value for: .

[0173] For example, for the extremely small feature "undervoltage protection response time," the original values ​​for the three candidate operating configurations are 0.08 seconds, 0.12 seconds, and 0.10 seconds, with a maximum value of 0.12 seconds. After forward processing, the resulting values ​​are 0.12 seconds. 0.08 = 0.04, 0.12 0.12 = 0, 0.12 0.10 = 0.02.

[0174] One possible implementation involves the computing device performing forward processing on intermediate target feature parameters using a distance transformation method based on the expected median value. Specifically, the computing device first obtains a preset expected median value for a certain intermediate feature parameter. (That is, the theoretical or standard value for optimal safety). Then, calculate the absolute deviation between the intermediate characteristic parameter and the expected intermediate value in each candidate running configuration. Next, identify the maximum absolute deviation of the target characteristic parameter across all candidate running configurations. Finally, based on the expected median value and maximum absolute deviation of this intermediate characteristic parameter, as shown in equation (21):

[0175] in, Let be the i target feature parameters, and be the intermediate target feature parameters.

[0176] For example, for the intermediate characteristic "grounding resistance value", assume its ideal safe intermediate value is 4 ohms. The original values ​​for the three candidate operating configurations are 3.5 ohms, 4.2 ohms, and 5.0 ohms, respectively. Their absolute deviations are 0.5, 0.2, and 1.0, respectively, with a maximum deviation M=1. After forward processing, the obtained values ​​are 1. 0.5 / 1.0=0.5, 1 0.2 / 1.0=0.8, 1 1.0 / 1.0=0.

[0177] For example, the process of a computing device standardizing a decision matrix can be represented as follows:

[0178] in, The parameter values ​​of the target feature parameters in the i-th row and j-th column are represented after standardization. This represents the parameter values ​​of the target feature parameters in the i-th row and j-th column after the forwarding process.

[0179] S402. By approximating the ideal solution and the decision matrix, determine the relative distance between each target feature parameter in each candidate running configuration and the corresponding target feature parameter in the desired running configuration.

[0180] The expected operating configurations include positive and negative expected operating configurations. A positive expected operating configuration is composed of the maximum parameter value in each column of the decision matrix, while a negative expected operating configuration is composed of the minimum parameter value in each column of the decision matrix. The relative distances include a first relative distance to the positive expected operating configuration and a second relative distance to the negative expected operating configuration.

[0181] Specifically, to quantify the difference between candidate operating configurations and the ideal reference point, the computing device calculates the absolute difference between the standardized value and the positive and negative expected values ​​for each candidate operating configuration and each feature, which are used as the first and second relative distance components, respectively. These components characterize the degree of deviation of the configuration from the optimal and worst states for each feature.

[0182] In some embodiments, the computing device can determine the desired operating configuration using a column-oriented extreme value extraction method. Specifically, the computing device traverses each column of the decision matrix. For each column, the computing device finds the maximum and minimum values, which are used as the values ​​for the positive and negative desired operating configurations on that feature, respectively. After traversal, the vector composed of the maximum values ​​of each column is the positive desired operating configuration, and the vector composed of the minimum values ​​of each column is the negative desired operating configuration.

[0183] For example, the process of obtaining the positive expected runtime configuration and the negative expected runtime configuration can be represented as follows:

[0184] in, This refers to the maximum value in each column of the decision matrix. This refers to the minimum value in each column of the decision matrix. m is the number of candidate running configurations (i.e., the total number of rows in the decision matrix).

[0185] In some embodiments, the computing device may employ the absolute distance method to calculate the relative distance between each candidate runtime configuration and the desired runtime configuration across each feature dimension.

[0186] For example, calculate the relative distance between the i-th candidate running configuration and the desired configuration ( The process can be represented as:

[0187] S403. Based on the first weight of each target feature parameter, perform weighted aggregation on the first relative distance of each target feature parameter in each candidate running configuration to obtain the first weighted distance from each candidate running configuration to the expected running configuration.

[0188] S404. Based on the first weight of each target feature parameter, the second relative distance of each target feature parameter in each candidate running configuration is weighted and aggregated to obtain the second weighted distance from each candidate running configuration to the negative expected running configuration.

[0189] The following explains S403 and S404: Specifically, the distance between each candidate operating configuration and the positive and negative expected safety states (i.e., the first relative distance) quantifies the merits of the configuration from both positive and negative perspectives. To reflect the differences in the impact of different features on safety, the computing device uses the weight of each target feature parameter (such as the first weight) to weight the distances across each feature dimension. By aggregating the weighted distances, a quantitative index (i.e., fit) is finally obtained that can comprehensively measure the overall gap between the candidate operating configuration and the ideal safety state.

[0190] One possible implementation is that the computing device can multiply the relative distance on each feature by its weight and then sum them to directly obtain the first or second weighted distance.

[0191] Another possible implementation involves the computing device first squaring the relative distances for each feature, then multiplying the squared result by the weight corresponding to that feature. Next, these products from all features are summed, and finally, the square root of the sum is taken to obtain a total weighted distance value. Applying this series of operations to positive distances using the first weight yields the first weighted distance; applying the same operation to negative distances using the second weight yields the second weighted distance. This method amplifies larger distance values ​​for individual features more significantly.

[0192] For example, the aggregation process using weighted Euclidean distance can be represented as follows:

[0193] in, This represents the first weight of the j-th target feature parameter (in some embodiments, it may also be the comprehensive weight mentioned above). , The distance between the j-th target feature parameter in the i-th candidate running configuration and the positive and negative expected running configurations. , This is the weighted distance between the i-th candidate running configuration and the positive and negative expected running configurations.

[0194] S405. Based on the first weighted distance and the second weighted distance of each candidate running configuration, determine the fit of each candidate running configuration.

[0195] Specifically, the first weighted distance represents the difference between the candidate configuration and the positive expected operating configuration (the smaller the better), and the second weighted distance represents the difference between the candidate configuration and the negative expected operating configuration (the larger the better). In order to obtain a unified evaluation index that comprehensively measures how well the candidate configuration is "close to the optimal and far from the worst", these two distance information need to be merged into a scalar value, namely the fit.

[0196] In some embodiments, the computing device may use a formula for calculating relative proximity to determine the degree of fit.

[0197] For example, the process can be represented as:

[0198] Essentially, it calculates the proportion of the relative distance from the candidate configuration to the negative ideal solution (worst-case state) in the total distance (the sum of the distances to the positive ideal solution and the distances to the negative ideal solution). Because The smaller the better. The bigger the better, therefore The closer the value is to 1, the closer the configuration is to the expected running configuration.

[0199] For example, candidate runtime configuration A =0.3, =0.7, then its fit is 0.7 / (0.3+0.7)=0.7. Candidate running configuration B. =0.3, =0.7, then its fit is 0.5 / (0.5+0.5)=0.5.

[0200] It should be understood that this scheme integrates multidimensional parameters into a unified evaluation scale by positively correcting heterogeneous safety indicators and weighted aggregation based on model weights. This enables candidate operational configurations to be ranked comparablely under a quantitative system that reflects true safety preferences, thereby ensuring the rationality of the optimal results and the effectiveness of the decision.

[0201] In an exemplary embodiment, this application also provides a flowchart illustrating another method for configuring the operation of an energy storage power station. For example... Figure 5 As shown, Characteristic parameters for the pre-configured operation of energy storage power stations: Define and specify the parameters used to describe and evaluate the safe operating status of energy storage power stations. These parameters constitute the basic dimensions for subsequent analysis and decision-making.

[0202] Based on the AdaBoost algorithm and the historical operating configurations of sample energy storage power stations, a configuration analysis model is trained: using the AdaBoost ensemble learning algorithm, with historical operating configuration data and its safety status results as samples, a machine learning model that can predict the safety status based on input features is trained.

[0203] Based on the training process data of the configuration analysis model, the importance quantification results of each feature parameter are determined: the internal data generated during the model training process (such as Gini gain value) are extracted and analyzed, and the importance of each feature parameter to the final safety status judgment of the model is calculated in a quantitative manner.

[0204] Based on the importance quantification results, the target feature parameters and their first weights are determined from the feature parameters: features are filtered according to their quantified importance, target feature parameters are selected, and objective weights (first weights) are assigned to them based on their importance quantification results.

[0205] Obtain candidate configuration parameters, process the candidate configuration parameters based on the analytic hierarchy process (AHP), and construct a judgment matrix: Obtain the specific parameter values ​​of multiple candidate running configurations to be evaluated, and use the AHP to compare the relative importance of each target feature parameter pairwise through expert experience or rules to form a judgment matrix.

[0206] Based on the judgment matrix, the second weight of the target feature parameters is determined: the constructed judgment matrix is ​​calculated to obtain the weight value of each target feature parameter based on subjective comparison, which is the second weight.

[0207] Based on the second weight and the judgment matrix, a consistency check is performed: the logic of pairwise comparisons in the judgment matrix is ​​checked for consistency in order to avoid self-contradictory evaluations.

[0208] At this point, the consistency check is performed. If it fails, it indicates a contradiction in the subjective judgment, and the judgment matrix needs to be corrected. If it succeeds, it indicates that the subjective weights are reliable, and the process continues.

[0209] Based on the second weight and the first weight of the target feature parameter, the comprehensive weight of the target feature parameter is obtained: the first weight objectively obtained by the model is combined with the second weight obtained based on subjective evaluation (such as weighted average) to obtain the final comprehensive weight of each target feature parameter.

[0210] The candidate configuration parameters are evaluated by combining the approximation ideal solution ranking method with comprehensive weights to obtain the fit between the candidate running configuration and the expected configuration: The approximation ideal solution ranking method is used to calculate the weighted distance between each candidate configuration and a virtual "optimal safe configuration" (positive ideal solution) and "worst safe configuration" (negative ideal solution) in the multi-dimensional feature space, thereby obtaining the fit score of each candidate configuration.

[0211] Based on the fit of candidate operating configurations, the target operating configuration is determined: the candidate operating configuration with the highest fit score is selected as the final recommended safe operating configuration for the target energy storage power station.

[0212] This safe operation configuration can be used to guide the operation, construction, and optimization of target energy storage power stations.

[0213] Specifically, Figure 5 For a detailed description of the steps described in the text, please refer to the above embodiments, which will not be elaborated here.

[0214] Figure 6 This is a schematic diagram of a processing device for the operation configuration of an energy storage power station, provided as an embodiment of this application. Figure 6 As shown, the processing unit configured for operation of this energy storage power station includes: Module 601 is used to acquire model training process data for the configuration analysis model. The configuration analysis model has the ability to output the safety status results of the energy storage power station based on multiple feature parameters in the energy storage power station's operational configuration. The configuration analysis model includes multiple base learners, and each base learner includes multiple nodes. The model training process data includes: the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner during feature splitting learning based on the feature parameters during model training.

[0215] Processing module 602 is used to determine the importance quantification result of each feature parameter based on the model training process data. The importance quantification result is used to characterize the degree of influence of the corresponding feature parameter on the safety status result of the energy storage power station output by the configuration analysis model. Based on the importance quantification result, target feature parameters are selected from the feature parameters, and the first weight of each target feature parameter is determined. The target feature parameter is the feature parameter whose importance quantification result satisfies the preset importance condition. The first weight is positively correlated with the degree of influence represented by the importance quantification result of the target feature parameter. Multiple candidate operating configurations of the target energy storage power station are obtained. Based on the first weight of each target feature parameter, the relative distance between each target feature parameter in each candidate operating configuration and the desired operating configuration is weighted and aggregated to determine the fit degree of each candidate operating configuration. The desired operating configuration is obtained based on the target feature parameters of multiple candidate operating configurations. The relative distance to the desired operating configuration is obtained by processing the target feature parameters of each candidate operating configuration using an approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate operating configuration and the desired operating configuration. The candidate operating configuration with the highest fit degree is determined from the candidate operating configurations and used as the target operating configuration of the target energy storage power station. The target operating configuration is used to configure the target energy storage power station. The safety status of the target energy storage power station after configuration is related to the target operating configuration.

[0216] In other embodiments, the characteristic parameters include at least one of the following dimensions: DC-side battery pack grounding characteristic dimension, power conversion characteristic dimension, relay protection characteristic dimension, and fault characteristic dimension. The DC-side battery pack grounding characteristic dimension includes: DC battery pack connection method parameters and grounding resistance value parameters. The power conversion characteristic dimension includes: converter operating condition parameters, converter performance parameters, and converter safety protection function parameters. The relay protection characteristic dimension includes: relay protection action criterion type parameters, relay protection area range parameters, relay protection device model parameters, and relay protection function parameters. The fault characteristic dimension includes: inter-pole fault current peak parameters, polar fault current peak parameters, positive and negative DC bus voltage peak parameters, DC energy storage capacitor fault voltage parameters, AC side three-phase current offset parameters, polar fault current response time parameters, undervoltage protection response time parameters, and converter blocking response time parameters.

[0217] In other embodiments, the processing module 602 is specifically configured to: for any one of the multiple feature parameters (first feature parameter), determine the sum of the Gini gain values ​​of the target node in each base learner from the model training process data. The target node is the node that performs feature splitting learning based on the first feature parameter. Based on the sum of the Gini gain values ​​corresponding to each base learner and the ensemble contribution weight of each base learner, determine the importance quantification result of the first feature parameter.

[0218] In other embodiments, the preset importance condition includes: the importance quantification result of the feature parameter is greater than a preset quantification result threshold. And / or, the importance quantification result of the feature parameter is ranked before a preset position in the order of importance quantification results from largest to smallest. The processing module 602 is specifically used to: normalize the importance quantification result of each target feature parameter and determine the first weight of each target feature parameter.

[0219] In other embodiments, the processing module 602 is further configured to: acquire relative importance assessment information for target feature parameters and construct a judgment matrix. The relative importance assessment information includes quantified values ​​of the relative importance of any two target feature parameters in the same dimension. The judgment matrix is ​​then processed by solving for normalized eigenvectors to obtain the second weight to be verified for each target feature parameter. Based on the second weight to be verified for each target feature parameter, a logical consistency check is performed on the judgment matrix to determine the logical consistency check result. If the logical consistency check result is passed, the second weight to be verified for each target feature parameter is determined as the second weight. Based on the first weight and the second weight of each target feature parameter, a comprehensive weight for each target feature parameter is determined. In this case, the processing module 602 is further configured to: based on the comprehensive weight of each target feature parameter, perform weighted aggregation of the relative distances between each target feature parameter in each candidate running configuration and the desired running configuration to determine the fit of each candidate running configuration.

[0220] In another possible implementation of the first aspect, the processing module 602 is specifically used to: perform positive and standardized processing on the parameter values ​​of each candidate running configuration for each target feature parameter, and construct a decision matrix based on each processed candidate running configuration. Each row in the decision matrix corresponds to a candidate running configuration, and each column corresponds to the parameter value of a target feature parameter. By approximating the ideal solution and the decision matrix, the relative distance between each target feature parameter in each candidate running configuration and the corresponding target feature parameter in the desired running configuration is determined. The desired running configuration includes: a positive desired running configuration and a negative desired running configuration. The positive desired running configuration is composed of the maximum parameter value in each column of the decision matrix, and the negative desired running configuration is composed of the minimum parameter value in each column of the decision matrix. The relative distance includes: a first relative distance with the positive desired running configuration and a second relative distance with the negative desired running configuration. Based on the first weight of each target feature parameter, the first relative distance of each target feature parameter in each candidate running configuration is weighted and aggregated to obtain the first weighted distance from each candidate running configuration to the positive desired running configuration. Based on the first weight of each target feature parameter, the second relative distance of each target feature parameter in each candidate running configuration is weighted and aggregated to obtain the second weighted distance from each candidate running configuration to the negative expected running configuration. Based on the first weighted distance and the second weighted distance of each candidate running configuration, the fit of each candidate running configuration is determined.

[0221] The energy storage power station operation configuration processing device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be found in the relevant descriptions in the method embodiment, and will not be repeated here. Furthermore, each module in the above energy storage power station operation configuration processing device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0222] Figure 7 This is a schematic diagram of the processing equipment configured for the operation of an energy storage power station, as provided in an embodiment of this application. Figure 7 As shown, the processing equipment configured for operation of this energy storage power station includes: a memory 701, a transceiver 702, and at least one processor 703.

[0223] The transceiver 702 is used to interact with other devices to send and receive data.

[0224] For example, in this embodiment of the application, transceiver 702 can be used to acquire model training process data of configuration analysis model.

[0225] The memory 701 stores computer program code, which includes computer instructions. These computer instructions run in the processing equipment configured for the operation of the aforementioned energy storage power station to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, or a USB flash drive, portable hard drive, read-only memory, magnetic disk, or optical disk, etc.

[0226] Processor 703 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Processor 703 can also be other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0227] The memory 701, transceiver 702, and processor 703 are communicatively connected. For example, the memory 701 and transceiver 702 can be connected to the processor 703 via a system bus to complete communication between them. The system bus can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, an industry standard architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus.

[0228] Optionally, the memory 701 can be either independent or integrated with the processor 703. When the memory 701 is configured independently, it is connected to the processor 703 via a system bus.

[0229] This application also provides a chip for executing instructions, which is used to execute the technical solution of the energy storage power station operation configuration processing method in the above embodiments.

[0230] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solution of the energy storage power station operation configuration processing method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the processing device for the energy storage power station operation configuration can execute the technical solution of the energy storage power station operation configuration processing method described in the above embodiments.

[0231] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the processing method for the operation configuration of the energy storage power station in the above embodiments.

[0232] The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0233] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0234] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0235] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0236] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0237] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0238] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0239] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0240] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for processing the operation configuration of an energy storage power station, characterized in that, The method includes: Acquire model training process data of the configuration analysis model; the configuration analysis model has the ability to output the safety status result of the energy storage power station based on multiple feature parameters in the operation configuration of the energy storage power station; the configuration analysis model includes multiple base learners, and each base learner includes multiple nodes; the model training process data includes: the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner when performing feature splitting learning based on the feature parameters during model training; Based on the model training process data, the importance quantification result of each feature parameter is determined; the importance quantification result is used to characterize the degree of influence of the corresponding feature parameter on the safety status result of the energy storage power station output by the configuration analysis model; Based on the importance quantification results, target feature parameters are selected from the feature parameters, and a first weight is determined for each target feature parameter; the target feature parameters are feature parameters whose importance quantification results satisfy a preset importance condition; the first weight is positively correlated with the degree of influence represented by the importance quantification results of the target feature parameters; Multiple candidate operating configurations of the target energy storage power station are obtained. Based on a first weight of each target feature parameter, the relative distances between each target feature parameter in each candidate operating configuration and the desired operating configuration are weighted and aggregated to determine the fit degree of each candidate operating configuration. The desired operating configuration is obtained based on the target feature parameters of the multiple candidate operating configurations. The relative distances with the desired operating configuration are obtained by processing the target feature parameters of each candidate operating configuration using an approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate operating configuration and the desired operating configuration. The candidate operating configuration with the highest fit is determined from the candidate operating configurations and used as the target operating configuration for the target energy storage power station; the target operating configuration is used to configure the target energy storage power station, and the safety status of the configured target energy storage power station is related to the target operating configuration.

2. The method according to claim 1, characterized in that, The feature parameters include at least one of the following dimensions: DC-side battery pack grounding characteristics, power conversion characteristics, relay protection characteristics, and fault characteristics; The grounding characteristics of the DC-side battery pack include: DC battery pack connection parameters and grounding resistance value parameters. The power conversion characteristic dimensions include: converter operating condition parameters, converter performance parameters, and converter safety protection function parameters. The relay protection feature dimensions include: relay protection action criterion type parameters, relay protection area range parameters, relay protection device model parameters, and relay protection function parameters; Fault characteristic dimensions include: peak parameters of inter-pole fault current, peak parameters of polar fault current, peak parameters of positive and negative DC bus voltage, fault voltage parameters of DC energy storage capacitor, AC side three-phase current offset parameters, polar fault current response time parameters, low voltage protection response time parameters, and converter blocking response time parameters.

3. The method according to claim 1, characterized in that, The determination of the importance quantification result of each feature parameter based on the model training process data includes: For any one of the plurality of feature parameters, the sum of the Gini gain values ​​of the target node in each base learner is determined from the model training process data; the target node is a node that performs feature splitting learning based on the first feature parameter. The importance quantification result of the first feature parameter is determined based on the sum of the Gini gain values ​​corresponding to each base learner and the ensemble contribution weight of each base learner.

4. The method according to claim 1, characterized in that, The preset importance conditions include: The importance quantization result of the feature parameter is greater than the preset quantization result threshold; And / or, The importance quantification results of the feature parameters are ranked before a preset position in the order of importance quantification results from largest to smallest. Determining the first weight for each of the target feature parameters includes: The importance quantification results of each target feature parameter are normalized to determine the first weight of each target feature parameter.

5. The method according to claim 4, characterized in that, After normalizing the importance quantification results of each target feature parameter and determining the first weight of each target feature parameter, the method further includes: Obtain relative importance assessment information for the target feature parameters and construct a judgment matrix; the relative importance assessment information includes the quantified values ​​of the relative importance of any two target feature parameters in the same dimension; The judgment matrix is ​​normalized and its eigenvectors are solved to obtain the second weight to be verified for each of the target feature parameters; Based on the second weight to be verified for each of the target feature parameters, a logical consistency test is performed on the judgment matrix to determine the logical consistency test result of the judgment matrix. If the logical consistency test result is passed, the second weight to be tested for each of the target feature parameters is determined as the second weight; Based on the first weight and the second weight of each target feature parameter, a comprehensive weight for each target feature parameter is determined; The step of weighted aggregation of the relative distances between each target feature parameter and the desired running configuration in each candidate running configuration, based on a first weight for each target feature parameter, to determine the fit of each candidate running configuration includes: Based on the comprehensive weight of each target feature parameter, the relative distance between each target feature parameter and the desired running configuration in each candidate running configuration is weighted and aggregated to determine the fit of each candidate running configuration.

6. The method according to claim 1, characterized in that, The process of obtaining multiple candidate operating configurations for the target energy storage power station, based on a first weight for each target feature parameter, involves weighted aggregation of the relative distances between each target feature parameter in each candidate operating configuration and the desired operating configuration to determine the fit of each candidate operating configuration, including: For each candidate running configuration, the parameter values ​​of each target feature parameter are positively oriented and standardized, and a decision matrix is ​​constructed based on each candidate running configuration after processing; each row of the decision matrix corresponds to a candidate running configuration, and each column corresponds to a parameter value of a target feature parameter; Using the approximation method and the decision matrix, the relative distance between each target feature parameter in each candidate running configuration and the corresponding target feature parameter in the desired running configuration is determined. The desired running configuration includes a positive desired running configuration and a negative desired running configuration. The positive desired running configuration is composed of the maximum parameter value in each column of the decision matrix, and the negative desired running configuration is composed of the minimum parameter value in each column of the decision matrix. The relative distance includes a first relative distance with the positive desired running configuration and a second relative distance with the negative desired running configuration. Based on the first weight of each of the target feature parameters, the first relative distance of each of the target feature parameters in each of the candidate running configurations is weighted and aggregated to obtain the first weighted distance from each candidate running configuration to the positive expected running configuration; Based on the first weight of each of the target feature parameters, the second relative distance of each of the target feature parameters in each of the candidate running configurations is weighted and aggregated to obtain the second weighted distance from each candidate running configuration to the negative expected running configuration; The fit of each candidate running configuration is determined based on the first weighted distance and the second weighted distance of each candidate running configuration.

7. A processing device for the operation configuration of an energy storage power station, characterized in that, include: Obtain the configuration data used to analyze the model training process. The configuration analysis model has the ability to output the safety status results of the energy storage power station based on multiple characteristic parameters in the operation configuration of the energy storage power station. The configuration analysis model includes multiple base learners, and each base learner includes multiple nodes; the model training process data includes: the ensemble contribution weight of each base learner and the Gini gain value generated by each node in each base learner when performing feature splitting learning based on the feature parameters during model training; The processing module is used to determine the importance quantification result of each feature parameter based on the model training process data; the importance quantification result is used to characterize the degree of influence of the corresponding feature parameter on the safety status result of the energy storage power station output by the configuration analysis model; based on the importance quantification result, select target feature parameters from the feature parameters and determine the first weight of each target feature parameter; the target feature parameter is a feature parameter whose importance quantification result satisfies a preset importance condition; the first weight is positively correlated with the degree of influence characterized by the importance quantification result of the target feature parameter; obtain multiple candidate operating configurations of the target energy storage power station, and based on the first weight of each target feature parameter, assign weights to each of the candidate operating configurations. The relative distances between the target feature parameters and the desired operating configuration are weighted and aggregated to determine the fit degree of each candidate operating configuration. The desired operating configuration is obtained based on the target feature parameters of the multiple candidate operating configurations. The relative distances to the desired operating configuration are obtained by processing the target feature parameters of each candidate operating configuration using the approximation ideal solution method. The fit degree characterizes the degree of fit between the corresponding candidate operating configuration and the desired operating configuration. The candidate operating configuration with the highest fit degree is determined from the candidate operating configurations and used as the target operating configuration of the target energy storage power station. The target operating configuration is used to configure the target energy storage power station, and the safety status of the configured target energy storage power station is related to the target operating configuration.

8. A processing device for the operation configuration of an energy storage power station, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, it causes the processing equipment configured for operation of the energy storage power station to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, When the computer program product is run on a computer / executed by the computer's processor, it implements the method as described in any one of claims 1-6.

Images